This study explores the multifaceted roles of generative AI in university students’ academic writing processes. While students often use AI to address writing challenges, understanding its impact on learning remains essential yet difficult to isolate. Providing effective guidance is crucial to fully harness the benefits of AI. Students typically use AI tools ad hoc and could often benefit from exploring more roles and improving their prompting skills. In this study we explore the experiences of students using AI tools beyond text generation focusing on the creative aspects of their academic work and explore emergent design requirements for AI tools. Data were gathered through a focused ethnography methodology during a workshop on AI tools for academic writing. Students display a positive but critical attitude towards AI tools, recognizing their potential while articulating criticisms regarding performance and reliability. They find AI-generated texts more reliable than images. Opinions on AI tools vary, with some students viewing them as helpful for brainstorming and time-saving, while others express concerns about overreliance and the need for critical evaluation. Some tools are particularly favored for their interface and control. Students emphasize the importance of effective guidance and task-specific tools. Emergent design requirements include transparency, usability, and structured views to enhance AI tool reliability and user experience.